Metadata Issue in 1.4.1

[SkillenUK]

Describe the bug I've moved from Ubuntu Server 12.04 (which I think ran Liquidsoap 1.1.1) to Ubuntu Server 20.04, which now has Liquidsoap 1.4.1 in the repository. I have never managed to get Liquidsoap installed in a way which I am happy with through OPAM or using the script so I stick to the version package manager, to be honest I was happy with 1.1.1.

I am running the same script which I ran in Liquidsoap 1.1.1 which worked as I expected it to. Now in Liquidsoap 1.4.1, when the server starts up the playlist is loaded and the metadata from the first file in my playlist comes through to the output.icecast object as expected.

When the track finishes and the playlist (autodj) moves to the next file, the metadata from the first track remains but another "REPLAY: " is appended to the start as the append_autodj(m) function is firing again. This continues this way, continuing to display the metadata for the first track with multiple appends at the start for every new track in the playlist.

If a live harbor connects then the correct metadata comes through from that, but when the live finished and the playlist (autodj) comes back on then the incorrect metadata from that first track returns. I have tried connecting through Telnet after it has been playing for a while (so has cycled through several tracks in the playlist) and the correct metadata can't be seen in any of the last played positions.

To Reproduce I've tried to provide a stripped down script containing the elements that could be causing the issue...

# Input harbor
djlive = input.harbor("live",port=8010,password="******")
djlivebuffer10 = input.harbor("livebuffer10",port=8010,password="******",buffer=10.,max=20.)

# Append live to the metadata
def append_live(m) =
  title = m["artist"]
  [("artist","LIVE: #{title}")]
end
djlive = map_metadata(append_live, djlive)
djlivebuffer10 = map_metadata(append_live, djlivebuffer10)

# Make safe then strip blank to give 10 seconds without auto DJ kicking in if the DJ drops out
live = strip_blank(max_blank=10.,mksafe(fallback(track_sensitive=false, [djlive,djlivebuffer10])))

# Auto dj playlist input
autodj = playlist(mode="randomize",reload=1,reload_mode="rounds","/etc/liquidsoap/autodj/autodj.m3u")

# Append replay to the metadata
def append_autodj(m) =
  title = m["artist"]
  [("artist","REPLAY: #{title}")]
end
autodj = map_metadata(append_autodj, autodj)

# Define transitions for changing between live and auto dj
def tolive(old,new) =
  sequence([fade.final(duration=1.,old),new])
end

def toauto(old,new)
  # Skip to next track on playlist
  source.skip(new)
  sequence(merge=true,[fade.initial(duration=2.,new)])
end

# Set the fallback between live and auto
testoutput = mksafe(fallback(track_sensitive=false, transitions=[tolive,toauto], [live,autodj]))

# Test transcoder : MP3 128 kbps stereo
output.icecast(
  %mp3(bitrate=128),
  mount="/test128",
  host="localhost", port=8000, password="******",
  name="Test Stream", description="128kbps MP3 stream", url="", genre="Music",
  testoutput)

Expected behavior When the fallback is playing the playlist (autodj), I would expect the metadata for the currently playing item to be passed through to the output.icecast object.

Version details

• OS: Ubuntu 20.04
• Version 1.4.1

Install method Ubuntu package manager (apt-get).

Common issues I have searched around, including in this issues log and I have been unable to find someone with the same issue. If you can point to an issue in my script or provide me with a workaround it would be good. I would not like to need to install a newer version through OPAM (it just seems messy that way). Hopefully the package in Ubuntu will be updated to a version which does not contain the issue soon. I am using the same script I had in v1.1.1.

[toots]

Hi! There's a good chance the issue is already fixed since 1.4.2: https://github.com/savonet/liquidsoap/issues/1115

[SkillenUK]

OK thanks, I've tackled OPAM again and managed to get 1.4.4 installed. The issue is resolved in that version. I can't seem to get the daemon script working with the OPAM install, I get the "Failed to start liquidsoap.service: Unit liquidsoap.service not found." message when I try to start the service after running the script.

[SkillenUK]

OK, so I'm still having the issue in 1.4.4, it appeared to be working at first but the issue is definately there. I have done some more testing and if I add a line in to mksafe the autodj object then output that directly then the metadata works. The issue is happening after someone has connected to one of the input.harbor objects. I have restarted the server and connected to liquidsoap via telnet then used the .skip command to play the next track and the metadata comes through correctly on all outputs. But after connection to the input.harbor, the autodj skips to the next track but the old metadata is retained on the output which has gone through the fallback routine. The append_autodj routine is running again and I am getting an extra "REPLAY: " added to the start. Maybe it's related to the transition where the command to skip to the next track is happening.

Now I've tested this some more I'm convinced that you will be able to reproduce the problem with the script above. Let me know if you need anything else, thanks.

[toots]

Ok, thanks for the effort. I'll try to reproduce.

[toots]

I can confirm the issue..

[toots]

@SkillenUK I would suggest to skip the track when leaving the playlist instead of entering it and also to do it in a asynchronous task in order to not confuse the switch:

# Define transitions for changing between live and auto dj
def tolive(old,new) =
  def skip() =
    source.skip(old)
    (-1.)
  end
  add_timeout(0., skip)
  sequence([fade.final(duration=1.,old),new])
end

def toauto(old,new)
  sequence(merge=true,[fade.initial(duration=2.,new)])
end

Let me know if this works for you.

[SkillenUK]

Thanks for your help. The correct metadata comes through with the solution above. On the first fallback I still get one extra append on the autodj source, i.e. 'REPLAY: REPLAY: